To celebrate the 10 millionth user Trello’s social media team implored their followers to tweet what they use @trello for, using the #trello10m hashtag. These tweets provide insights into consumer use patterns which this analysis will summarize. I outline these insights here, first detailing the sample of Trello followers. I then show the global geography of the Trello tweetosphere, before utilizing natural language processing techniques to summarize Trello twitter followers’ self-reported use patterns as indicated by their tweets.

Tweetosphere Sample

I received the tweets second hand via the social media team, who used the Tweetbinder to collect the tweets. This data provides rudimentary description of the tweetosphere along with the text of the tweets. Unfortunately a few of the files overlapped tweet samples. I removed these duplicate tweets (i.e. tweets with the same text, user, and posting time) along with tweets from official Trello accounts, I then used several resources–including Google’s Chromium Compact Language Detector–to identify the language of the tweets to filter out non-English tweets in the text analysis.

Note: Given the short character length of tweets a small share of tweets were falsely identified as English, but may remain in the “English” sample. These were later removed and thus didn’t effect the textual analysis.

Geography of Tweetosphere

The global reach of the #trello10m tweet is depicted below. To create the map I geocoded each tweeters’ self identified location using OpenStreetMap. The map is displayed at a global level but will cluster and aggregate to the corresponding geography as you zoom in and out. If you zoom all the way in to a location (i.e. Brooklyn) each tweet is represented by a point and the text of the tweet is displayed if you click on the tweet’s marker.

Note: Ideally the tweets would be geocoded via the latitude and longitude coordinates of the tweet itself, unfortunately the sample coverage of these coordinates is sparse. As a fallback I used the tweeters’ self identified location to geocode the tweets. Because a small number of users will sporadically input locations like “Candyland” or “Antarctica”, this data isn’t preferred over the geocodes of the tweets themselves, however given the large sample size of the tweetosphere and random distribution of these false locations this methodology should provide an accurate overview of the geography of the #trello10m tweetosphere. I also limited the influence of these false locations by only accepting actual place names (i.e. not businesses, streets, etc.) in the OpenStreetMap database.

Textual Analysis

Arguably the most useful information from the campaign can be gleamed from the tweets themselves. Followers were asked to respond what they used Trello for and provided template tweets such as, ‘I use @trello for [unicorn spotting adventures] .’, to guide them. I analyzed these tweets and summarize the findings below.

Note: Some users simply left the tweet templates as is, these tweets were removed from the analysis. I also left one template tweet in ‘[herding sheep]’ because followers chose to add to this tweet.

Word Frequencies

The first stage of the analysis consisted of a simple summary of word occurrences. As expected terms like organize, manage, and plan dominate trello use patterns.

Other Uses

Frequency Table

Phrase Frequencies

A more detailed analysis of the co-occurrence of words provides insights into exactly what Trello users organize, manage, and plan.

Word Association & Textual Analysis

To examine these uses in more detail clusters allows the formation of broader themes to emerge. Below is a textual network of the most common co-occurring words. The plot can been zoomed in and individual nodes can be highlighted by clicking on them.